Applied AI Systems Engineer · Multi-Agent LLM Research · Decision Intelligence MS Data Science @ George Washington University — Expected May 2026
I build production-oriented AI systems that go beyond model outputs — focusing on reliability, failure modes, and decision-aware coordination across agents, pipelines, and retrieval layers.
Agentis — Open-source multi-agent AI platform Deploy fleets of specialized agents across 12 LLM providers. Watch them think, collaborate, and deliver in real time. → github.com/Dhwanil25/Agentis
Multi-Agent Coordination Research Studying failure modes in LLM agent systems using El Farol-style environments — where agents must coordinate without communication and emergent behaviour determines outcomes.
HPC AI Sandbox GPU-accelerated model orchestration using Slurm — secure pipelines, parallel inference, and reproducible evaluation at scale.
RAIN Decision intelligence platform for retail inventory management and supplier risk — built on hybrid retrieval and structured LLM outputs.
- Multi-agent LLM coordination and failure taxonomy
- Hybrid retrieval — BM25 + vector search + reranking
- LLM guardrails and structured JSON outputs
- HPC-based model orchestration (Slurm + GPU)
- Evaluation pipelines and monitoring before deployment
AI systems fail in subtle ways — and most people only notice after deployment.
I work backwards from failure: identifying breakdown points, designing evaluation before building, and treating LLMs as probabilistic reasoning engines rather than deterministic truth sources.
📍 Virginia, United States
